add dumbest possible 32-bit SkJumper backend

Everything uses a ton of stack, nothing tail calls, and for now this is
non-Windows only.  But, it does run faster than the portable serial code.

On my trashcan, running `monobench SkRasterPipeline_compile`:
   - Normal 64-bit AVX build:  43.6ns
   - Before this CL, 32-bit:  707.9ns
   - This CL:                 147.5ns

Change-Id: I4a8929570ace47193ed8925c58b70bb22d6b1447
Reviewed-on: https://skia-review.googlesource.com/20964
Reviewed-by: Mike Reed <reed@google.com>
Reviewed-by: Herb Derby <herb@google.com>
Commit-Queue: Mike Klein <mtklein@chromium.org>
3 files changed